Improving ΔΔG Predictions with a Multitask Convolutional Siamese Network.

Discovery of highly potent phosphodiesterase-1 inhibitors by a combined-structure free energy perturbation approach

Understanding and predicting the energetics of protein-protein interactions is fundamental to the structural modeling of protein complexes. Binding free energy can be approximated as a sum of pairwise atomic or residue contact energies, which are commonly inferred from contact frequencies observed in experimental protein structures. However, such statistically inferred potentials require certain assumptions and approximation. Here, we explore the possibility of deriving atomic and residue contact potentials directly from experimental binding free energy changes following mutation and present a number of such potentials. The first set of potentials is obtained by unweighted least-squares fitting and bootsrap aggregating. The second set is calculated using a weighting scheme optimized against absolute binding affinity data, so as to account for the over-representation of certain complexes, residues, and families of interactions. The congruence of the potentials with known physical chemistry is investigated. The potentials are further validated by ranking and clustering protein-protein docking poses.

An approach using the finite difference solution of the Poisson‐Boltzmann equation to estimate binding free energy changes for two receptor–ligand systems, arabinose binding protein and sulfate binding protein, is presented. The eight calculated binding free energy changes agree with experiment, showing a correlation coefficient of 0.92 and energy deviations of 1 kcal/mol or less. More importantly, the decomposition of solvation and assembly energies in this approach provides an understanding of binding mechanisms and therefore could suggest directions to alter binding affinities. The method is demonstrated to be useful in analyzing experimental binding structures and predicting binding effects of mutants or modified ligands for macromolecular systems, in which the electrostatic forces dominate the overall interaction and the structural perturbations upon modifications are small. © 1995 by John Wiley & Sons, Inc.

AlphaFold 3 (AF3), the latest version of protein structure prediction software, goes beyond its predecessors by predicting protein-protein complexes. It could revolutionize drug discovery and protein engineering, marking a major step toward comprehensive, automated protein structure prediction. However, independent validation of AF3's predictions is necessary. In this work, we evaluate AF3 complex structures using the SKEMPI 2.0 database which involves 317 protein-protein complexes and 8338 mutations. AF3 complex structures when applied to the most advanced TDL model, MT-TopLap (MultiTask-Topological Laplacian), give rise to a very good Pearson correlation coefficient of 0.86 for predicting protein-protein binding free energy changes upon mutation, which is slightly less than the 0.88 achieved earlier with the Protein Data Bank (PDB) structures. Nonetheless, AF3 complex structures led to a 8.6% increase in the prediction RMSE compared to original PDB complex structures. Additionally, some of AF3's complex structures have large errors, which were not captured in its ipTM performance metric. Finally, it is found that AF3's complex structures are not reliable for intrinsically flexible regions or domains.

The fast evolution of SARS-CoV-2 and other infectious viruses poses a grand challenge to the rapid response in terms of viral tracking, diagnostics, and design and manufacture of monoclonal antibodies (mAbs) and vaccines, which are both time-consuming and costly. This underscores the need for efficient computational approaches. Recent advancements, like topological deep learning (TDL), have introduced powerful tools for forecasting emerging dominant variants, yet they require deep mutational scanning (DMS) of viral surface proteins and associated three-dimensional (3D) protein-protein interaction (PPI) complex structures. We propose an AlphaFold 3 (AF3)-assisted multi-task topological Laplacian (MT-TopLap) strategy to address this need. MT-TopLap combines deep learning with topological data analysis (TDA) models, such as persistent Laplacians (PL) to extract detailed topological and geometric characteristics of PPIs, thereby enhancing the prediction of DMS and binding free energy (BFE) changes upon virus mutations. Validation with four experimental DMS datasets of SARS-CoV-2 spike receptor-binding domain (RBD) and the human angiotensin-converting enzyme-2 (ACE2) complexes indicates that our AF3 assisted MT-TopLap strategy maintains robust performance, with only an average 1.1% decrease in Pearson correlation coefficients (PCC) and an average 9.3% increase in root mean square errors (RMSE), compared with the use of experimental structures. Additionally, AF3-assisted MT-TopLap achieved a PCC of 0.81 when tested with a SARS-CoV-2 HK.3 variant DMS dataset, confirming its capability to accurately predict BFE changes and adapt to new experimental data, thereby showcasing its potential for rapid and effective response to fast viral evolution.

Predicting the effects of protein and DNA mutations on the binding free energy of protein-DNA complexes is crucial for understanding how DNA variants impact wild-type cellular function. As many cellular interactions involve protein-DNA binding, accurately predicting changes in binding free energy (ΔΔG) is valuable for distinguishing pathogenic mutations from benign ones. This study describes the development and optimization of the SAMPDI-3Dv2 machine learning method, which is trained on an expanded database of experimentally measured ΔΔGs. This enhanced model incorporates new features, including the 3D structure of the mutant protein, features of the mutant structure, and a position-specific scoring matrix (PSSM). Benchmarking was conducted using 5-fold cross-validation. The updated SAMPDI-3D model (SAMPDI-3Dv2) achieved Pearson correlation coefficients (PCCs) of 0.68 for protein and 0.80 for DNA mutations. These results represent significant improvements over existing tools. Additionally, the method's rapid execution time enables genome-scale predictions. The improved SAMPDI-3Dv2 shows enhanced predictive performance for analyzing mutations in protein-DNA complexes. By leveraging structural information and an expanded training dataset, SAMPDI-3Dv2 provides researchers with a more accurate and efficient tool for mutation analysis, contributing to identifying pathogenic variants and improving our understanding of cellular function.

Alchemical free energy simulations are amongst the most accurate techniques for the computation of the free energy changes associated with noncovalent protein-ligand interactions. A procedure is presented to estimate the relative binding free energies of several ligands to the same protein target where multiple, low-energy configurational substates might coexist, as opposed to one unique structure. The contributions of all individual substates were estimated, explicitly, with the free energy perturbation method, and combined in a rigorous fashion to compute the overall relative binding free energies and dissociation constants. It is shown that, unless the most stable bound forms are known a priori, inaccurate results may be obtained if the contributions of multiple substates are ignored. The method was applied to study the complex formed between human catechol-O-methyltransferase and BIA 9-1067, a newly developed tight-binding inhibitor that is currently under clinical evaluation for the therapy of Parkinson's disease. Our results reveal an exceptionally high-binding affinity (K(d) in subpicomolar range) and provide insightful clues on the interactions and mechanism of inhibition. The inhibitor is, itself, a slowly reacting substrate of the target enzyme and is released from the complex in the form of O-methylated product. By comparing the experimental catalytic rate (k(cat)) and the estimated dissociation rate (k(off)) constants of the enzyme-inhibitor complex, one can conclude that the observed inhibition potency (K(i)) is primarily dependent on the catalytic rate constant of the inhibitor's O-methylation, rather than the rate constant of dissociation of the complex.

The electrostatically embedded generalized molecular fractionation with conjugate caps (EE-GMFCC) method has been successfully utilized for efficient linear-scaling quantum mechanical (QM) calculation of protein energies. In this work, we applied the EE-GMFCC method for calculation of binding affinity of Endonuclease colicin-immunity protein complex. The binding free energy changes between the wild-type and mutants of the complex calculated by EE-GMFCC are in good agreement with experimental results. The correlation coefficient (R) between the predicted binding energy changes and experimental values is 0.906 at the B3LYP/6-31G*-D level, based on the snapshot whose binding affinity is closest to the average result from the molecular mechanics/Poisson-Boltzmann surface area (MM/PBSA) calculation. The inclusion of the QM effects is important for accurate prediction of protein-protein binding affinities. Moreover, the self-consistent calculation of PB solvation energy is required for accurate calculations of protein-protein binding free energies. This study demonstrates that the EE-GMFCC method is capable of providing reliable prediction of relative binding affinities for protein-protein complexes. © 2018 Wiley Periodicals, Inc.

We describe an efficient method for the simultaneous solution of all free energies within a relative binding free-energy (RBFE) network with cycle closure and experimental/reference constraint conditions using Bennett Acceptance Ratio (BAR) and Multistate BAR (MBAR) analysis. Rather than solving the BAR or MBAR equations for each transformation independently, the simultaneous solution of all transformations are obtained by performing a constrained minimization of a global objective function. The nonlinear optimization of the objective function is subjected to affine linear constraints that couple the free energies between the network edges. The constraints are used to enforce the closure of thermodynamic cycles within the RBFE network, and to enforce an additional set of linear constraint conditions demonstrated here to be subsets of (1 or 2) experimental values. We describe details of the practical implementation of the network BAR/MBAR procedure, including use of generalized coordinates in the minimization of the free-energy objective function, propagation of bootstrap errors from those coordinates, and performance and memory optimization. In some cases it is found that use of restraints in the optimization is more practical than use of generalized coordinates for enforcing constraint conditions. The fast BARnet and MBARnet methods are used to analyze the RBFEs of six prototypical protein-ligand systems, and it is shown that enforcement of cycle closure conditions reduces the error in the predictions only modestly, and further reduction in errors can be achieved when one or two experimental RBFEs are included in the optimization procedure. These methods have been implemented into FE-ToolKit, a new free-energy analysis toolkit. The BARnet/MBARnet framework presented here opens the door to new, more efficient and robust free-energy analysis with enhanced predictive capability for drug discovery applications.

Pneumonia remains a leading cause of mortality among children under five years of age, accounting for approximately 1.6 million deaths annually. Early detection is critical for reducing child mortality; however, traditional diagnostic methods rely on trained personnel and medical resources, often scarce in low-resource settings. Machine learning has emerged as a promising approach to pneumonia detection, but its effectiveness is hindered by the limited availability of labeled data required for training robust models. This study reviews existing machine learning techniques for pneumonia classification, including convolutional neural networks (CNNs), transfer learning, and few-shot learning approaches. Furthermore, we propose an optimized few-shot learning model that integrates Siamese networks with transfer learning to improve pneumonia detection using minimal labeled data. The model leverages MobileNetV3 as a pre-trained feature extractor, producing high-quality embeddings that enhance similarity learning within the Siamese framework. Additionally, triplet loss is incorporated to ensure a more discriminative embedding space, facilitating robust classification. Experimental results indicate that our model outperforms conventional machine learning classifiers, achieving an accuracy of 92.04%, precision of 91.20%, recall of 90.32%, and F1-score of 90.09%. The integration of Siamese networks with triplet loss enhances generalization while mitigating overfitting, making it particularly suitable for resource-constrained environments. These findings highlight the effectiveness of few-shot learning in pneumonia detection, offering a computationally efficient and scalable solution that bridges critical gaps in medical imaging.

We present an extension of the alchemical transfer method (ATM) for the estimation of relative binding free energies of molecular complexes applicable to conventional, as well as scaffold-hopping, alchemical transformations. Named ATM-RBFE, the method is implemented in the free and open-source OpenMM molecular simulation package and aims to provide a simpler and more generally applicable route to the calculation of relative binding free energies than what is currently available. ATM-RBFE is based on sound statistical mechanics theory and a novel coordinate perturbation scheme designed to swap the positions of a pair of ligands such that one is transferred from the bulk solvent to the receptor binding site while the other moves simultaneously in the opposite direction. The calculation is conducted directly in a single solvent box with a system prepared with conventional setup tools, without splitting of electrostatic and nonelectrostatic transformations, and without pairwise soft-core potentials. ATM-RBFE is validated here against the absolute binding free energies of the SAMPL8 GDCC host-guest benchmark set and against protein-ligand benchmark sets that include complexes of the estrogen receptor ERα and those of the methyltransferase EZH2. In each case the method yields self-consistent and converged relative binding free energy estimates in agreement with absolute binding free energies and reference literature values, as well as experimental measurements.

Exploring the medicinal significance of bioactive compounds through computational methods is an increasingly practiced approach in contemporary medicinal research. This study aims to assess the antidiabetic potential of compounds extracted from the plant Nyctanthes arbor tristis by evaluating their ability to inhibit the carbohydrate metabolic enzyme α-glucosidase. The research work was conducted through molecular docking calculation, molecular dynamics simulation (MDS), and ADMET prediction techniques. Among the compounds, arbortistoside-C (NAS03), and arbortristoside-D (NAS04) found in the seed of the plant were identified as hit inhibitors of the target protein with docking scores, -9.9 and -9.4 kcal/mol, respectively. The compounds showed a comparable docking score with the drug of diabetes acarbose (-8.6 kcal/mol). Geometrical parameters like radius of gyration, solvent accessibility surface, root mean square deviation, and root mean square fluctuation from MDS supported the stability of the protein-ligand complex. MMPBSA calculations demonstrated the stability and feasibility of the complex with binding free energy changes of -29.06±6.06 and -23.58±8.80 kcal/mol for compounds NAS03 and NAS04, respectively. The ADMET prediction suggested the drug-likeness of the compounds compared with that of the standard drugs. The results could be used in proposing the antidiabetic potential of the two compounds from the plant as a potential inhibitors of α-glucosidase enzyme. Further, in vitro and in vivo experiments on such compounds could be a more reliable path to validate the output of this computational research.

Family 19 chitinase from Aeromonas sp. No.10S-24 (72.6 kDa) is composed of two chitin-binding domains (ChBDs), two proline- and threonine-rich (PT-rich) linkers, and a catalytic domain. The purified enzyme was labile in a standard buffer condition and spontaneously degraded into a 46-kDa fragment upon storage at 4 degrees C. The N-terminal sequence of the 46-kDa fragment was found to correspond to the sequence of the C-terminal region of the second PT-rich linker, indicating that the 46-kDa fragment is produced by truncation of the two ChBDs and the two PT-rich linkers from the mature protein, and consists only of the catalytic domain. The hydrolytic activities toward insoluble and soluble substrates were significantly reduced by the truncation of two ChBDs. In addition, antifungal activity determined from the digestion rate of haustoria of powdery mildew was reduced by the ChBD truncation. Although the profile of the time-course of N-acetylglucosamine hexasaccharide [(GlcNAc)6] degradation catalyzed by the ChBD-truncated enzyme was similar to that of the mature enzyme protein, the specific activity of the ChBD-truncated enzyme determined from the rate of hexasaccharide degradation was lower than that of the mature enzyme. The two CBDs appear to be responsible for facilitating the hydrolytic reaction. The sugar residue affinities (binding free energy changes) at the individual subsites, (-2) (-1) (+1) (+2) (+3) (+4), were estimated by modeling the hexasaccharide hydrolysis by the mature and ChBD-truncated enzymes. The truncation of ChBDs was found to strongly affect the affinity at the (-1) site. This situation seems to result in the lower enzymatic activity of the ChBD-truncated enzyme toward the chitinous substrates.

We present calculations of the absolute and relative binding free energies of complexation of streptavidin with biotin and its analogs by means of a thermodynamic free energy perturbation method implemented with molecular dynamics. Using the recently solved crystal structure of the streptavidin-biotin complex, biotin was mutated into a dummy molecule as well as thiobiotin and iminobiotin both in the protein and in solution. The calculated absolute binding free energy was dependent on the simulation model used. Encouragingly, the "best models" provided a reasonable semiquantitative reproduction (-20 to -22 kcal/mol) of the experimental free energy (-18.3 kcal/mol). Furthermore, the calculated results give clear insights into the binding nature of the protein-ligand complex, showing that the van der Waals energy dominates the electrostatic and hydrogen bonding energies in the binding of biotin by streptavidin. Specifically, the mutation of biotin into a dummy molecule in solution has a delta G (van der Waals) approximately -4 kcal/mol, due to the cancellation of dispersion and repulsion "cavity" effects. On the other hand, in the protein, a very small free energy price must be paid to create a cavity since one already exists and the mutation of biotin into a dummy molecule has a delta G (van der Waals) approximately 15 kcal/mol. These results are also consistent with the interpretation that the entropy increase to be expected from hydrophobic interactions from desolvation of biotin is counterbalanced by a decrease in entropy accompanying the formation of buried hydrogen bonds, which have been derived from the apparently conflicting experimental data. They provide an alternative interpretation of the reason for the extremely high affinity of the biotin-streptavidin interaction than that recently proposed by Weber et al. (J. Am. Chem. Soc. 114:3197, 1992). In the case of the relative binding free energies, the calculated values of 3.8 +/- 0.6 and 7.2 +/- 0.6 kcal/mol compare well with the experimental values of 3.6 and 6.2 kcal/mol for the perturbation of biotin to thiobiotin and iminobiotin, respectively in the related protein avidin. The calculations indicate that desolvation of the ligand is important in understanding the relative affinity of the ligands with the protein. The above successful simulations suggest that the molecular dynamics/free energy perturbation method is useful for understanding the energetic features affecting the binding between proteins and ligands, since it is generally difficult to determine these factors unambiguously by experiment.(ABSTRACT TRUNCATED AT 400 WORDS)

BackgroundNatural language processing (NLP) has become an emerging technology in health care that leverages a large amount of free-text data in electronic health records to improve patient care, support clinical decisions, and facilitate clinical and translational science research. Recently, deep learning has achieved state-of-the-art performance in many clinical NLP tasks. However, training deep learning models often requires large, annotated data sets, which are normally not publicly available and can be time-consuming to build in clinical domains. Working with smaller annotated data sets is typical in clinical NLP; therefore, ensuring that deep learning models perform well is crucial for real-world clinical NLP applications. A widely adopted approach is fine-tuning existing pretrained language models, but these attempts fall short when the training data set contains only a few annotated samples. Few-shot learning (FSL) has recently been investigated to tackle this problem. Siamese neural network (SNN) has been widely used as an FSL approach in computer vision but has not been studied well in NLP. Furthermore, the literature on its applications in clinical domains is scarce.ObjectiveThe aim of our study is to propose and evaluate SNN-based approaches for few-shot clinical NLP tasks.MethodsWe propose 2 SNN-based FSL approaches, including pretrained SNN and SNN with second-order embeddings. We evaluate the proposed approaches on the clinical sentence classification task. We experiment with 3 few-shot settings, including 4-shot, 8-shot, and 16-shot learning. The clinical NLP task is benchmarked using the following 4 pretrained language models: bidirectional encoder representations from transformers (BERT), BERT for biomedical text mining (BioBERT), BioBERT trained on clinical notes (BioClinicalBERT), and generative pretrained transformer 2 (GPT-2). We also present a performance comparison between SNN-based approaches and the prompt-based GPT-2 approach.ResultsIn 4-shot sentence classification tasks, GPT-2 had the highest precision (0.63), but its recall (0.38) and F score (0.42) were lower than those of BioBERT-based pretrained SNN (0.45 and 0.46, respectively). In both 8-shot and 16-shot settings, SNN-based approaches outperformed GPT-2 in all 3 metrics of precision, recall, and F score.ConclusionsThe experimental results verified the effectiveness of the proposed SNN approaches for few-shot clinical NLP tasks.